Advancing polypharmacology-based drug discovery using deep generative models

In drug discovery, compounds with activity against more than one target protein due to target-specific interactions (multi-specificity), must be clearly distinguished from those with undesired promiscuity resulting from non-specific binding or artifacts (such as colloidal aggregators or other assay interference compounds). The former multi-target compounds (MT-CPDs) are desirable as polypharmacological agents because they are capable of intervening in altered pharmacological networks that play a role in various disease states. Since the early to mid-2000s, the polypharmacology paradigm has complemented (but not replaced) the concept of target-specific compounds for therapeutic interventions that had dominated drug discovery research since the 1980s as a result of biological reductionism. Notably, compound multi-specificity can also be therapeutically undesirable, as it may cause adverse side effects of drugs directed against a specific target. While most MT-CPDs are discovered in biological screening campaigns or profiling assays, they have also become an important topic in drug design. In addition to the use of conventional computational approaches, there have been increasing attempts since the early 2020s to design MT-CPDs with deep generative models, as discussed herein.